Memory with Memory in Tree-Based Genetic Programming

In recent work on linear register-based genetic programming (GP) we introduced the notion of Memory-with-Memory (MwM), where the results of operations are stored in registers using a form of soft assignment which blends a result into the current content of a register rather than entirely replace it. The MwM system yielded very promising results on a set of symbolic regression problems. In this paper, we propose a way of introducing MwM style behaviour in treebased GP systems. The technique requires only very minor modifications to existing code, and, therefore, is easy to apply. Experiments on a variety of synthetic and real-world problems show that MwM is very beneficial in tree-based GP, too.